Length: seconds, frames, and which unit is honest
The length column holds the clip length a repository states, in the unit the repository chose. One family states up to sixty seconds; another states a frame count and translates it itself. Frames and seconds do not convert without a frame rate nobody published. As of 2026-09-12.
| Family | Stated length | The unit, and what it is for |
|---|---|---|
| CogVideoX | Up to 161 frames on CogVideoX1.5-5B | Frames, translated by the vendor as 5 or 10 seconds |
| HunyuanVideo | 129 frames, as a measurement setting | Frames, sizing a memory figure rather than a ceiling |
| LTX-Video | Up to 60 seconds on the 13B model | Seconds, from the release notes |
| Mochi 1 | No length recorded | The repository states a present resolution instead |
| Wan | A five-second 720P clip | Seconds, as the worked example behind a timing figure |
Inclusion rule. Clip lengths a repository states for a released model, kept in the published unit. A length used only to describe a measurement is marked as such rather than read as a ceiling. Order. Alphabetical by family name.
1Frames and seconds are not the same claim
CogVideoX states up to 161 frames on CogVideoX1.5-5B and translates that itself as five or ten seconds, which only works because a frame rate decides which. LTX-Video states support for up to sixty seconds on its 13B model. The column keeps whichever unit the vendor published and does not turn one into the other.
Converting would mean choosing a frame rate the repository never stated, and the converted figure would then sit in the cell looking exactly like a quoted one. Frames are the precise unit and seconds are the readable one; where a vendor supplies both, so does the register.
2A supported length is a statement about the code
Sixty seconds means the pipeline accepts sixty seconds. It is not a claim that a face stays the same face for a minute, that the camera behaves, or that anyone would use the result, and none of those are things a repository is in a position to state.
Nothing here comes from running a model, so the cell stops where the repository stops. What it does establish is the ceiling a plan may even attempt, which is the one part of this that cannot be discovered by trying harder on a shorter clip.
3The longest ceiling sits on the side with no price
The sixty-second figure belongs to published weights. The seconds actually on sale are billed per generation and carry rates instead: LTX Studio lists LTX-2.5 from $0.09 to $0.39 a second, and Wan 3.0 is sold from about $0.06 at 480P. Neither rate attaches to the release that states the ceiling.
Wan's own repository states length a third way, as a five-second 720P clip used to time a run on a consumer card. That is a worked example behind a throughput figure, and filing it as a limit would be the register inventing a constraint out of a demonstration. The two sides are set against each other on clip length.
- LengthThe repository states support for up to 60 seconds of video on LTXV-13Bfrom the release notes
- Length and resolutionCogVideoX1.5-5B states 1360x768 and up to 161 frames, described as 5 or 10 secondsper model
- HardwareThe repository states TI2V-5B runs on consumer-grade graphics cards such as a 4090, generating a 5-second 720P video in under 9 minutes on a single consumer GPUvendor's own figure
- What the hosted service sellsWan 3.0 seconds at published per-second rates, from about $0.06 at 480P
- What the hosted service sellsLTX-2.5 API seconds at $0.09 to $0.39
4Sources
Values in this column are quoted from the LTX-Video repository, the LTX Studio API page, read 2026-09-12, and from the repositories linked on each model page. The other columns are listed on fields; what counts as a published statement is on how read. Next column: Licence, Size.